Noisy version of super-resolution (sr_bicubic) and Gaussian deblurring (deblur_gauss) experiments

[NebiyouTen]

Hello, thanks for releasing the DDNM code.

I ran into a couple of issues when trying the noisy versions of the SRx4 and deblurring tasks:

1. For deblur_gauss, using --add_noise and --sigma_y 0.05 gives very poor results. The PSNR drops below 10, which seems unexpected. It might be that the noise isn't being handled correctly?

2. For sr_bicubic, using --add_noise raises a NotImplementedError saying that SRConv does not implement the lambda function. It looks like the noisy version might not be supported in this case.

Could you please let me know whether these noisy settings are expected to work?

Thanks again!

[wuyudeyu]

Hello, I am currently implementing the deblurring task under DDNM+denoising. With the add_noise command, adding a slight amount of sigma-y noise results in psnr below 10, and the lpips result is also not satisfactory. How did you solve this deblurring denoising task? I am currently conducting a comparative experiment and do not know how to compare the deblurring denoising task of DDNM+. If you can solve this problem, can you help me? Thank you very much

[NebiyouTen]

Hello, I ended up with only comparing with DDNM is the noiseless setting. I did not compare with comparative experiment DDNM+.